Garment Manufacturing and Women’s Work, Reproduction, and Human Capital

class: center, middle, title-slide

.title[
# Garment Manufacturing and Women’s Work, Reproduction, and Human Capital
]
.subtitle[
## SM Shihab Siddiqui
]
.author[
### University of Oregon
]
.date[
### December, 2022
]

---

# Introduction

Since 1960s, developing countries experienced:

--
- Large .hi[declines in fertility]

- Improvements in .hi[women's education]

Can .hi[industrialization] friendly towards .hi[women's employment] explain these changes?

`$\uparrow$` Opportunity costs in child-bearing and child-rearing

`$\uparrow$` Returns to human capital in agrarian settings

`$\implies$` Fertility reduction and increased human capital accumulation

.footnote[
[1] *Ager et al (2020), Guldi and Rahman (2022), and Brown and Guinnane (2018)* shows that a similar dynamic played out during the second industrial revolution in US and Europe.]

---
# Introduction

Bangladesh started industrialization in 1980s through .hi[Garment manufacturing]
  
--

- First .hi[non-farm employment opportunity] for millions of women
  
--

- In a context of high fertility, low education, and  non-coercive family planning program
  
--

Allows me to examine the salience of industrialization `$\rightarrow$` `$\uparrow$` women's employment `$\rightarrow$` `$\downarrow$` fertility & `$\uparrow$` human capital accumulation of women.

Specifically, I test the .hi[effect of women's employment opportunities in garment manufacturing on:]

- **Female labor force participation (FLFP)**
- Age-specific marriage and fertility rates
- School enrollment, literacy, and years of schooling

---
# Why do we care?

.hi[Manufacturing- and export-led growth and development] is important. Especially relevant since:

- Pre-mature de-industrialization in developing countries (Rodrik, 2016).

--
  
  - Prevalence of women in garments industry is decreasing in Bangladesh.<sup>1<sup>
  
--

Adds to the literature focusing on .hi[trade and lives of workers] *(Autor et al (2013), Atkin (2016), Li (2018) and Autor et al (2019))*.

.footnote[ [1] Along the lines of what happens as technology improves in a manufacturing sector (Tejani and Kucera, 2021).]
---
# Related work

.hi[Heath and Mobarak (2015, JDE)] finds that Bangladeshi garments industry:

- Increased FLFP substantially, delayed marriage, and reduced fertility; and reduced women's education gap.

- Individual data is from a 2009 survey of 1395 households in sixty villages in four regions with garments.

- Relatively large distance between villages near and far from garments. Estimated treatment  effect may reflect spatial pattern of development.

- Factory data is from a 2014 survey of local individuals.

In contrast, I use .hi[census data], and .hi[construct a detailed factory-level dataset] of nearly all exporting garment factories across all regions of Bangladesh between 1991-2006.

---
# Preview of the paper

## Methods

I estimate the long run effect of employment opportunities on FLFP, reproductive, and human capital accumulation choices by:

- Using shift-share methods from the trade literature
 
--
  
 - Specifically, my .hi[identifying variation comes from differences in product specialization] within the garment industry only in regions that had garments in them.

---
# Preview of the paper

## Results

Employment opportunities in garment industry led to:

- .hi[Substantial increases in FLFP], and especially industrial FLFP

- .hi[Small decline in school] enrollment

- No statistically significant changes in literacy rates or years of schooling
  
--

- No changes in age-specific marriage and fertility rates.

---
# Organization of the paper

The rest of the presentation will:

- Discuss the context and economic question in greater detail

- Present the methodology and data

- Present the findings

- Conclude the paper

---
class: inverse, middle, center

# The Garments Industry and Women's Employment Oppurtunity

---
# Garments industry expansion

.pull-left[
<img src="data:image/png;base64,#jmp_pres1_files/figure-html/unnamed-chunk-1-1.svg" width="100%" style="display: block; margin: auto;" />
]

.pull-right[

- <font size="5">  Grew at .hi[11% per year] since 1991.

]

.pull-right[

- Second largest exporter of garments in the world, and accounts for .hi[85% of Bangladesh's exports] in recent decades.

]

.pull-right[

- Accounted for about .hi[33% of industrial employment] in 2017.

</font>
]
---
# Knit versus woven prodcuts

Ready made garments account for .hi[more than 90%] of exports

Two broad categories of products:
--

.pull-left[

.hi[Knit: HS code 61]

Single yarn looped repeatedly

Examples: Most sweaters, cotton T-shirts
]

--
.pull-right[

.hi[Woven: HS code 62]

Multiple yarn criss-crossed over and under each other

Examples: Shirts, jackets, pants.
]
---
# .smaller[Knit versus woven specialization]

- Producing woven is more energy and capital intensive, and commands about 10% higher per unit price (Sytsma, 2022).

- Woven factories are larger, and employs more women.

- Labor tasks overlap.
---
# Garments and FLFP

## A long history

Textile and affiliated industry always employed relatively more women across different time and place.

- Mid 1800s England (Burnette, 2008), USA (Field-Hendrey, 1998); developed and developing countries 1981-2008 (Kucera and Tejani, 2014).
  
--
  
- Women engaged in more spinning and knitting for centuries *(Virginia Postrel, Textiles and the Fabric of Civilization)*
  
--

- Surveys find that as high as .hi[60% of workers] in exporting garments industry are women.

---
# .smaller[Against the grain: Bangladeshi FLFP]

- Among 20-24 year olds, FLFP was about 49% in 2015 (ADB 2016)

---
# Conceptual framework

.hi[Recap:] Substantial expansion of an exporting industry in the industrialization process that is favorable to women's work.

--
  - Clear prediction: .hi[FLFP will increase]

Theoretically, FLFP, fertility, marriage, and schooling decisions are endogenous.

- Labor market opportunities changes net benefits of marriage and fertility *(see Greenwood et al (2017) for a review).*

- Human capital accumulation could increase or decrease depending on skill intensity of exports *(Atkin (2016), Li (2018))*

- In turn, effects fertility along the Quality-Quantity trade-off.
  
--

.hi[What happened to reproductive and human capital accumulation choices is an empirical questions!]

---
class: inverse, middle, center

# Empirical strategy

---
# Research question

The goal is to estimate the long-term effect of women's employment opportunities in garment manufacturing on:

- Female labor force participation (FLFP)

- Age-specific marriage and fertility rates

- School enrollment, literacy, and years of schooling

---
# Data sources

- .hi[Outcome and control data] is obtained by aggregating individual-level data from the Bangladesh Census 1991 (10% sub-sample), 2001 (10% sub-sample), and 2011 (5% sub-sample) to sub-districts.

- I created a novel .hi[Factory-level dataset] by combining four different datasets from Bangladesh Garments Manufacturing and Exporters Association (BGMEA) datasets with:

--
  - Factory locations, date of establishment, number of machines in the factory, and the type of product they produce (knit versus woven, or mixed).
  
--

- Trade data from Comtrade (2022)

---
# Identification challenge

## Why not simple OLS?

$$
`\begin{equation}
  Y_{s,t} = \beta \space \text{Female Labor Demand}_{s,t} + Z_{s,t} \beta_z + v_{s,t}
\end{equation}`
$$
--

- `$Y_{s,t}$` is an outcome of interest in sub-district `$s$` at decade ending in year `$t$`.
- `$Z_{s,t}$` is a vector of controls

Two key issues:

- .hi[Correlation between garment location and development] of a sub-district

- Labor demand, or production, or exports from a sub-district is .hi[not directly observed]

---
# Spatial spread of garments

<div class="figure" style="text-align: center">
<img src="data:image/png;base64,#figures/rmgspread.png" alt="Spread of garment factories between 1991-2006" width="75%" />
<p class="caption">Spread of garment factories between 1991-2006</p>
</div>

- Factory locations dependent on .hi[infrastructure quality] (kagy, 2014).

- I also verify that proxies of infrastructure conditions explain much more of the variation in existence of factories in a sub-district.

---
# .smaller[Garments and development]

Location of garment factory closely follows the spatiotemporal pattern of development.

I solve the endogeneity from omitted measures of development .hi[using several steps.]

---
# First difference model

`\begin{equation}
  \Delta Y_{s,t} = \beta \space \Delta \text{Female Labor Demand}_{s,t} + \delta_{t}  + Z_{s,t-10} \beta_z + X_{s,t-10} \beta_x + \epsilon_{s,t}
\end{equation}`

- `$\Delta Y_{s,t}$` is the .hi[decadal change] in outcome variables in sub-district `$s$` over decade ending at year `$t$`.

- `$\delta_t$` are .hi[census-year fixed-effects]

- `$Z_{s, t-10}$` is a vector of .hi[start of decade values of controls common in all regressions:]

- Three proxies of .hi[infrastructure conditions:] electrification rate, urbanization rate, and density
  
  - Two measures of .hi[demographic conditions:] share working age (15-64) population, and average years of education of working age population.
  
---
# First difference model

`\begin{equation}
  \Delta Y_{s,t} = \beta \space \Delta \text{Female Labor Demand}_{s,t} + \delta_{t}  + Z_{s,t-10} \beta_z + X_{s,t-10} \beta_x + \epsilon_{s,t}
\end{equation}`

- `$X_{s,t-10}$` .hi[start of decade outcomes for males] other than in the cases of regressions corresponding to marriage and fertility rates.

## Sample restriction

Sub-districts with and without factories are very different.

- Sample period of 1991-2011 overlaps with large changes education, urbanization, and infrastructure.

- .hi[I restrict the analysis only to sub-districts that had a factory by 2006]

---
# Measuring labor demand

Taking inspiration from Autor (2013), one candidate measure of `$\Delta \text{Female Labor Demand}_{s,t}$` is:

`\begin{equation}
  \sum_{i=0}^{9} \alpha_{s,t-i}^{K} * \frac{\text{Export}_{BD,t-i}^{K}}{L_{s,t-i}} + \sum_{i=1}^{9} \alpha_{s,t-i}^{W} * \frac{\text{Export}_{BD,t-i}^{W}}{{L_{s,t-i}}} \\
 \alpha_{s,t-i}^{K} = \frac{Machines_{s,t-i}^{K}}{Machines_{BD,t-i}^{K}}, 
\alpha_{s,t-i}^{W} = \frac{Machines_{s,t-i}^{W}}{Machines_{BD,t-i}^{W}}
\end{equation}`

- Apportions .hi[total knit (woven) exports] originating in Bangladesh .hi[to a sub-district based on that sub-district's share of national knit (woven) production capacity].

- Scales by population

I call this measure `$\Delta \text{Export Exposure}_{s,t}$`.
---
# Export exposure

I use beginning of the decade values of:

- `$\alpha_{s,t-10}^{K}$`, `$\alpha_{s,t-10}^{W}$`, and `$L_{t-10}$`
  
to account for location-time specific omitted shocks. Thus,

`\begin{align}
  \Delta \text{ Export Exposure}_{s,t} =& \alpha_{s,t-10}^{K} * \frac{\Delta \space \text{Export}_{BD,t}^{K}}{L_{t-10}} + \alpha_{s,t-10}^{W} * \frac{\Delta \space \text{Export}_{BD,t}^{W}}{L_{t-10}} 
\end{align}`

---
# Regression model

The regression model is:

`\begin{equation}
  \Delta Y_{s,t} = \beta \space \Delta \text{Export Exposure}_{s,t} + \delta_{t}  + Z_{s,t-10} \beta_z + X_{s,t-10} \beta_x + \epsilon_{s,t}
\end{equation}`

---
# Identifying assumption

Adopting Goldsmith-Pinkham et al (2020), the key assumption is that the differences in knit and woven specialization do not *change* outcomes through confounders. That is:

- .hi[Identifying assumption 1:] Extent of knit versus woven specialization in a sub-district is uncorrelated with the errors conditional on controls.

- .hi[Identifying assumption 2:] FLFP responds similarly to woven and knit shocks.

- .hi[Example of a violation:]	Woven employs more women, so areas with more woven factories has increased presence of fertility control programs because there are more women.
---

class: inverse, middle, center

# Factory Data
---
# Data challenge

Need to estimate `$\alpha_{s,1991}^{K}$`, `$\alpha_{s,2001}^{K}$`, and `$\alpha_{s,1991}^{W}, \alpha_{s,2001}^{K}$`

- Need data on factory establishment date, location, number of machines **in 1991 and 2001**, and whether they are a knit or woven factory.
  
--

- Created new dataset with estimated number of machines in 1991 and 2001 using BGMEA directory 2000-01, and 2009-10

- Matched using unique BGMEA number, names, addresses, and phone numbers.

- Matched 87% of factories in 2000-01 directory
---
# Two stylized factory facts

- .hi[Factory life-cycles:] virtually no exit till early 2000s, exits starts gaining steam between 2000-2010.

- Factory exits are unlikely to lead to erroneous share estimates.

---
# Two stylized factory facts

<div class="figure" style="text-align: center">
<img src="data:image/png;base64,#figures/facSizeDist.png" alt="Factory Size Distribution" width="70%" />
<p class="caption">Factory Size Distribution</p>
</div>

- Firm size distribution seems constant over 1991-2011 except at the top end. Suggestive of no major technological change.
---
class: inverse, middle, center

# Results
---
# Correlates of Shares

<div class="figure" style="text-align: center">
<p class="caption">Correlates of Shares</p><img src="data:image/png;base64,#figures/tab_conf.png" alt="Correlates of Shares" width="60%" /></div>

- .hi[Shares in 91 and 2001 is not correlated with subsequent changes.]

---
# Garments and FLFP

`\begin{equation}
  \Delta Y_{s,t} = \beta \space \Delta \text{Export Exposure}_{s,t} + \delta_{t}  + Z_{s,t-10} \beta_z + X_{s,t-10} \beta_x + \epsilon_{s,t}
\end{equation}`

Outcomes I consider are `$\Delta FLFPR_{s,t}$` and `$\Delta FLFPR\text{-Ind}_{s,t}$` for:

- Working age women (Ages 15-64)

- Women Ages 15-29 (most common in garments industry)

- Women Ages 15-20 (has schooling and marriage implications)

`$\Delta \text{Export exposure}_{s,t}$` is .hi[$1,000 per working age person] in a decade.
---
# Garments and overall FLFP

- Mean exposure of $854 in 2 decades, .hi[about 6-12 months of average income].

- `$\uparrow$` Overall FLFP by .hi[3.47 - 5.36 percentage points]

--
  - Roughly .hi[32% of the increases in FLFP] in areas with garment factories over 1991-2011.
  
---
# Garments and Industrial-FLFP

- Mean exposure of $854 in 2 decades, .hi[about 6-12 months of average income].

- `$\uparrow$` Overall FLFP by .hi[4.58 - 6.46 percentage points]

--
  - Roughly .hi[55% of the increases in industrial-FLFP] in areas with garment factories over 1991-2011.

---
# .smaller[Discussions: Garments and FLFP]

- Garments industry is a major contributor to increases in FLFP.

- Greater effect on industrial FLFP

- Greater effect for younger women.

- Using a DiD estimator, .hi[Heath and Mobarak (2015)] found that women in villages within commuting zones of garment factories had about 15 percentage point more FLFP.

- My magnitude is about half.

---
# .smaller[Discussions: Garments and FLFP]

- Garment factories are in urban regions

- With .hi[50 percent] urbanization in my sample sub-districts, FLFP increases could be as high as  7-10.5 percentage points in a sample similar to Heath and Mubarak (2015)

- .hi[Possibly, the effect of garments on Bangladeshi FLFP is smaller than  previously thought, though it is quite substantial.]

---
# .smaller[Discussions: Garments and FLFP]

Autor et al. (2019) estimates suggest two decades of Chinese import competition reduced manufacturing employment as a share of population by 2.12 percentage points.

- My estimate of effect of export exposure on FLFP is much larger, reflecting the relatively more important role of garment industry in the Bangladeshi FLFP context.
---
# Garments and schooling

`\begin{equation}
  \Delta Y_{s,t} = \beta \space \Delta \text{Export Exposure}_{s,t} + \delta_{t}  + Z_{s,t-10} \beta_z + X_{s,t-10} \beta_x + \epsilon_{s,t}
\end{equation}`

Outcomes I consider are:

- `$\Delta \text{Enrollment}_{s,t}$` of girls:

- Ages 5-9, Ages 10-13 (tells us about changes in attitudes towards women's education)
  
--

- Ages 14-19 (tells us about trade-offs in FLFP and education)
  
--

- `$\Delta \text{Literacy}_{s,t}$` and `$\Delta \text{Years of Schooling}_{s,t}$` of girls ages 14-19 (tells us about overall effect)

---
# Garments and enrollment

- At the mean exposure value, enrollment of working age girls (14-19 year old) .hi[reduces by 2.71 percentage point]

- No effect in enrollment of other ages.
---
# .smaller[Garments and human capital accumulation]

- No effect on literacy

- Not surprising, literacy is established earlier in school

- Average years of schooling is negative and almost statistically significant **(t-value is -1.625)**

---
# .smaller[Garments and human capital accumulation]

- Slight `$\downarrow$` in human capital accumulation

- .hi[Contradicts] the findings of Heath and Mubarak (2015)

- Findings in line with:

--
  
  - Atkin (2016): Export manufacturing opportunity reduced school enrollment of older teenagers in 1980-2005
  
--

- Li (2018): Expansion of low (high) low-skill (high-skill) exports reduces (increases) schooling

---
# .smaller[Garments and reproductive behavior]

<div class="figure" style="text-align: center">
<img src="data:image/png;base64,#figures/marfert.png" alt="Age specific ever-married and realized fertility rates" width="70%" />
<p class="caption">Age specific ever-married and realized fertility rates</p>
</div>

---
# .smaller[Garments and reproductive behavior]

`\begin{equation}
  \Delta Y_{s,t} = \beta \space \Delta \text{Export Exposure}_{s,t} + \delta_{t}  + Z_{s,t-10} \beta_z + \epsilon_{s,t}
\end{equation}`

Outcomes I consider are:

- `$\Delta \text{Marriage Rates}_{s,t}$` for 15-20, and 21-30 years old (ages when employment at garments is most likely)

- `$\Delta \text{Fertility Rates}_{s,t}$` for 21-30 (period of `$\uparrow$` opportunity cost of child-bearing and reading), and 30-40 years old (tests changes in timing of child-bearing).

---
# Garments and marriage

<div class="figure" style="text-align: center">
<img src="data:image/png;base64,#figures/tab_marr.png" alt="Age specific ever-married and realized fertility rates" width="70%" />
<p class="caption">Age specific ever-married and realized fertility rates</p>
</div>

- No impact on marriage rates

---
# Garments and fertility

- No impact on fertility rates

---
# .smaller[Discussion: Garments and reproductive behavior]

- No effect on marriage and fertility behavior

- Heath and Mobarak (2015) finds very small effects in case of Bangladesh.

- Autor (2019) finds small effects in case of US.

- Bangladesh was going through a very fast fertility transition already.

---
# Conclusion

Bangladesh started its industrialization in late 1970s.

- The garment industry played a dominant role. Moreover,

- Millions of Bangladeshi women got their first non-farm employment opportunity.

---
# Conclusions

- I found that this increased opportunity did increase FLFP substantially

- Reduced schooling for older girls, but without serious effect on overall human capital accumulation

- No effect on fertility

At least in the case of Bangladesh, and perhaps more generally, ongoing industrialization may not explain the rapid fertility transition and improvements in women's education.

---
class: clear, center, middle

# Thanks